net/sched/act_api.c

   1 /*
   2  * net/sched/act_api.c  Packet action API.
   3  *
   4  *              This program is free software; you can redistribute it and/or
   5  *              modify it under the terms of the GNU General Public License
   6  *              as published by the Free Software Foundation; either version
   7  *              2 of the License, or (at your option) any later version.
   8  *
   9  * Author:      Jamal Hadi Salim
  10  *
  11  *
  12  */
  13
  14 #include <linux/types.h>
  15 #include <linux/kernel.h>
  16 #include <linux/string.h>
  17 #include <linux/errno.h>
  18 #include <linux/slab.h>
  19 #include <linux/skbuff.h>
  20 #include <linux/init.h>
  21 #include <linux/kmod.h>
  22 #include <linux/err.h>
  23 #include <linux/module.h>
  24 #include <linux/rhashtable.h>
  25 #include <linux/list.h>
  26 #include <net/net_namespace.h>
  27 #include <net/sock.h>
  28 #include <net/sch_generic.h>
  29 #include <net/pkt_cls.h>
  30 #include <net/act_api.h>
  31 #include <net/netlink.h>
  32
  33 static int tcf_action_goto_chain_init(struct tc_action *a, struct tcf_proto *tp)
  34 {
  35         u32 chain_index = a->tcfa_action & TC_ACT_EXT_VAL_MASK;
  36
  37         if (!tp)
  38                 return -EINVAL;
  39         a->goto_chain = tcf_chain_get_by_act(tp->chain->block, chain_index);
  40         if (!a->goto_chain)
  41                 return -ENOMEM;
  42         return 0;
  43 }
  44
  45 static void tcf_action_goto_chain_fini(struct tc_action *a)
  46 {
  47         tcf_chain_put_by_act(a->goto_chain);
  48 }
  49
  50 static void tcf_action_goto_chain_exec(const struct tc_action *a,
  51                                        struct tcf_result *res)
  52 {
  53         const struct tcf_chain *chain = a->goto_chain;
  54
  55         res->goto_tp = rcu_dereference_bh(chain->filter_chain);
  56 }
  57
  58 static void tcf_free_cookie_rcu(struct rcu_head *p)
  59 {
  60         struct tc_cookie *cookie = container_of(p, struct tc_cookie, rcu);
  61
  62         kfree(cookie->data);
  63         kfree(cookie);
  64 }
  65
  66 static void tcf_set_action_cookie(struct tc_cookie __rcu **old_cookie,
  67                                   struct tc_cookie *new_cookie)
  68 {
  69         struct tc_cookie *old;
  70
  71         old = xchg((__force struct tc_cookie **)old_cookie, new_cookie);
  72         if (old)
  73                 call_rcu(&old->rcu, tcf_free_cookie_rcu);
  74 }
  75
  76 /* XXX: For standalone actions, we don't need a RCU grace period either, because
  77  * actions are always connected to filters and filters are already destroyed in
  78  * RCU callbacks, so after a RCU grace period actions are already disconnected
  79  * from filters. Readers later can not find us.
  80  */
  81 static void free_tcf(struct tc_action *p)
  82 {
  83         free_percpu(p->cpu_bstats);
  84         free_percpu(p->cpu_qstats);
  85
  86         tcf_set_action_cookie(&p->act_cookie, NULL);
  87         if (p->goto_chain)
  88                 tcf_action_goto_chain_fini(p);
  89
  90         kfree(p);
  91 }
  92
  93 static void tcf_action_cleanup(struct tc_action *p)
  94 {
  95         if (p->ops->cleanup)
  96                 p->ops->cleanup(p);
  97
  98         gen_kill_estimator(&p->tcfa_rate_est);
  99         free_tcf(p);
 100 }
 101
 102 static int __tcf_action_put(struct tc_action *p, bool bind)
 103 {
 104         struct tcf_idrinfo *idrinfo = p->idrinfo;
 105
 106         if (refcount_dec_and_lock(&p->tcfa_refcnt, &idrinfo->lock)) {
 107                 if (bind)
 108                         atomic_dec(&p->tcfa_bindcnt);
 109                 idr_remove(&idrinfo->action_idr, p->tcfa_index);
 110                 spin_unlock(&idrinfo->lock);
 111
 112                 tcf_action_cleanup(p);
 113                 return 1;
 114         }
 115
 116         if (bind)
 117                 atomic_dec(&p->tcfa_bindcnt);
 118
 119         return 0;
 120 }
 121
 122 int __tcf_idr_release(struct tc_action *p, bool bind, bool strict)
 123 {
 124         int ret = 0;
 125
 126         /* Release with strict==1 and bind==0 is only called through act API
 127          * interface (classifiers always bind). Only case when action with
 128          * positive reference count and zero bind count can exist is when it was
 129          * also created with act API (unbinding last classifier will destroy the
 130          * action if it was created by classifier). So only case when bind count
 131          * can be changed after initial check is when unbound action is
 132          * destroyed by act API while classifier binds to action with same id
 133          * concurrently. This result either creation of new action(same behavior
 134          * as before), or reusing existing action if concurrent process
 135          * increments reference count before action is deleted. Both scenarios
 136          * are acceptable.
 137          */
 138         if (p) {
 139                 if (!bind && strict && atomic_read(&p->tcfa_bindcnt) > 0)
 140                         return -EPERM;
 141
 142                 if (__tcf_action_put(p, bind))
 143                         ret = ACT_P_DELETED;
 144         }
 145
 146         return ret;
 147 }
 148 EXPORT_SYMBOL(__tcf_idr_release);
 149
 150 static size_t tcf_action_shared_attrs_size(const struct tc_action *act)
 151 {
 152         struct tc_cookie *act_cookie;
 153         u32 cookie_len = 0;
 154
 155         rcu_read_lock();
 156         act_cookie = rcu_dereference(act->act_cookie);
 157
 158         if (act_cookie)
 159                 cookie_len = nla_total_size(act_cookie->len);
 160         rcu_read_unlock();
 161
 162         return  nla_total_size(0) /* action number nested */
 163                 + nla_total_size(IFNAMSIZ) /* TCA_ACT_KIND */
 164                 + cookie_len /* TCA_ACT_COOKIE */
 165                 + nla_total_size(0) /* TCA_ACT_STATS nested */
 166                 /* TCA_STATS_BASIC */
 167                 + nla_total_size_64bit(sizeof(struct gnet_stats_basic))
 168                 /* TCA_STATS_QUEUE */
 169                 + nla_total_size_64bit(sizeof(struct gnet_stats_queue))
 170                 + nla_total_size(0) /* TCA_OPTIONS nested */
 171                 + nla_total_size(sizeof(struct tcf_t)); /* TCA_GACT_TM */
 172 }
 173
 174 static size_t tcf_action_full_attrs_size(size_t sz)
 175 {
 176         return NLMSG_HDRLEN                     /* struct nlmsghdr */
 177                 + sizeof(struct tcamsg)
 178                 + nla_total_size(0)             /* TCA_ACT_TAB nested */
 179                 + sz;
 180 }
 181
 182 static size_t tcf_action_fill_size(const struct tc_action *act)
 183 {
 184         size_t sz = tcf_action_shared_attrs_size(act);
 185
 186         if (act->ops->get_fill_size)
 187                 return act->ops->get_fill_size(act) + sz;
 188         return sz;
 189 }
 190
 191 static int tcf_dump_walker(struct tcf_idrinfo *idrinfo, struct sk_buff *skb,
 192                            struct netlink_callback *cb)
 193 {
 194         int err = 0, index = -1, s_i = 0, n_i = 0;
 195         u32 act_flags = cb->args[2];
 196         unsigned long jiffy_since = cb->args[3];
 197         struct nlattr *nest;
 198         struct idr *idr = &idrinfo->action_idr;
 199         struct tc_action *p;
 200         unsigned long id = 1;
 201
 202         spin_lock(&idrinfo->lock);
 203
 204         s_i = cb->args[0];
 205
 206         idr_for_each_entry_ul(idr, p, id) {
 207                 index++;
 208                 if (index < s_i)
 209                         continue;
 210
 211                 if (jiffy_since &&
 212                     time_after(jiffy_since,
 213                                (unsigned long)p->tcfa_tm.lastuse))
 214                         continue;
 215
 216                 nest = nla_nest_start(skb, n_i);
 217                 if (!nest) {
 218                         index--;
 219                         goto nla_put_failure;
 220                 }
 221                 err = tcf_action_dump_1(skb, p, 0, 0);
 222                 if (err < 0) {
 223                         index--;
 224                         nlmsg_trim(skb, nest);
 225                         goto done;
 226                 }
 227                 nla_nest_end(skb, nest);
 228                 n_i++;
 229                 if (!(act_flags & TCA_FLAG_LARGE_DUMP_ON) &&
 230                     n_i >= TCA_ACT_MAX_PRIO)
 231                         goto done;
 232         }
 233 done:
 234         if (index >= 0)
 235                 cb->args[0] = index + 1;
 236
 237         spin_unlock(&idrinfo->lock);
 238         if (n_i) {
 239                 if (act_flags & TCA_FLAG_LARGE_DUMP_ON)
 240                         cb->args[1] = n_i;
 241         }
 242         return n_i;
 243
 244 nla_put_failure:
 245         nla_nest_cancel(skb, nest);
 246         goto done;
 247 }
 248
 249 static int tcf_del_walker(struct tcf_idrinfo *idrinfo, struct sk_buff *skb,
 250                           const struct tc_action_ops *ops)
 251 {
 252         struct nlattr *nest;
 253         int n_i = 0;
 254         int ret = -EINVAL;
 255         struct idr *idr = &idrinfo->action_idr;
 256         struct tc_action *p;
 257         unsigned long id = 1;
 258
 259         nest = nla_nest_start(skb, 0);
 260         if (nest == NULL)
 261                 goto nla_put_failure;
 262         if (nla_put_string(skb, TCA_KIND, ops->kind))
 263                 goto nla_put_failure;
 264
 265         idr_for_each_entry_ul(idr, p, id) {
 266                 ret = __tcf_idr_release(p, false, true);
 267                 if (ret == ACT_P_DELETED) {
 268                         module_put(ops->owner);
 269                         n_i++;
 270                 } else if (ret < 0) {
 271                         goto nla_put_failure;
 272                 }
 273         }
 274         if (nla_put_u32(skb, TCA_FCNT, n_i))
 275                 goto nla_put_failure;
 276         nla_nest_end(skb, nest);
 277
 278         return n_i;
 279 nla_put_failure:
 280         nla_nest_cancel(skb, nest);
 281         return ret;
 282 }
 283
 284 int tcf_generic_walker(struct tc_action_net *tn, struct sk_buff *skb,
 285                        struct netlink_callback *cb, int type,
 286                        const struct tc_action_ops *ops,
 287                        struct netlink_ext_ack *extack)
 288 {
 289         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 290
 291         if (type == RTM_DELACTION) {
 292                 return tcf_del_walker(idrinfo, skb, ops);
 293         } else if (type == RTM_GETACTION) {
 294                 return tcf_dump_walker(idrinfo, skb, cb);
 295         } else {
 296                 WARN(1, "tcf_generic_walker: unknown command %d\n", type);
 297                 NL_SET_ERR_MSG(extack, "tcf_generic_walker: unknown command");
 298                 return -EINVAL;
 299         }
 300 }
 301 EXPORT_SYMBOL(tcf_generic_walker);
 302
 303 static bool __tcf_idr_check(struct tc_action_net *tn, u32 index,
 304                             struct tc_action **a, int bind)
 305 {
 306         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 307         struct tc_action *p;
 308
 309         spin_lock(&idrinfo->lock);
 310         p = idr_find(&idrinfo->action_idr, index);
 311         if (IS_ERR(p)) {
 312                 p = NULL;
 313         } else if (p) {
 314                 refcount_inc(&p->tcfa_refcnt);
 315                 if (bind)
 316                         atomic_inc(&p->tcfa_bindcnt);
 317         }
 318         spin_unlock(&idrinfo->lock);
 319
 320         if (p) {
 321                 *a = p;
 322                 return true;
 323         }
 324         return false;
 325 }
 326
 327 int tcf_idr_search(struct tc_action_net *tn, struct tc_action **a, u32 index)
 328 {
 329         return __tcf_idr_check(tn, index, a, 0);
 330 }
 331 EXPORT_SYMBOL(tcf_idr_search);
 332
 333 bool tcf_idr_check(struct tc_action_net *tn, u32 index, struct tc_action **a,
 334                    int bind)
 335 {
 336         return __tcf_idr_check(tn, index, a, bind);
 337 }
 338 EXPORT_SYMBOL(tcf_idr_check);
 339
 340 int tcf_idr_delete_index(struct tc_action_net *tn, u32 index)
 341 {
 342         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 343         struct tc_action *p;
 344         int ret = 0;
 345
 346         spin_lock(&idrinfo->lock);
 347         p = idr_find(&idrinfo->action_idr, index);
 348         if (!p) {
 349                 spin_unlock(&idrinfo->lock);
 350                 return -ENOENT;
 351         }
 352
 353         if (!atomic_read(&p->tcfa_bindcnt)) {
 354                 if (refcount_dec_and_test(&p->tcfa_refcnt)) {
 355                         struct module *owner = p->ops->owner;
 356
 357                         WARN_ON(p != idr_remove(&idrinfo->action_idr,
 358                                                 p->tcfa_index));
 359                         spin_unlock(&idrinfo->lock);
 360
 361                         tcf_action_cleanup(p);
 362                         module_put(owner);
 363                         return 0;
 364                 }
 365                 ret = 0;
 366         } else {
 367                 ret = -EPERM;
 368         }
 369
 370         spin_unlock(&idrinfo->lock);
 371         return ret;
 372 }
 373 EXPORT_SYMBOL(tcf_idr_delete_index);
 374
 375 int tcf_idr_create(struct tc_action_net *tn, u32 index, struct nlattr *est,
 376                    struct tc_action **a, const struct tc_action_ops *ops,
 377                    int bind, bool cpustats)
 378 {
 379         struct tc_action *p = kzalloc(ops->size, GFP_KERNEL);
 380         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 381         int err = -ENOMEM;
 382
 383         if (unlikely(!p))
 384                 return -ENOMEM;
 385         refcount_set(&p->tcfa_refcnt, 1);
 386         if (bind)
 387                 atomic_set(&p->tcfa_bindcnt, 1);
 388
 389         if (cpustats) {
 390                 p->cpu_bstats = netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
 391                 if (!p->cpu_bstats)
 392                         goto err1;
 393                 p->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
 394                 if (!p->cpu_qstats)
 395                         goto err2;
 396         }
 397         spin_lock_init(&p->tcfa_lock);
 398         p->tcfa_index = index;
 399         p->tcfa_tm.install = jiffies;
 400         p->tcfa_tm.lastuse = jiffies;
 401         p->tcfa_tm.firstuse = 0;
 402         if (est) {
 403                 err = gen_new_estimator(&p->tcfa_bstats, p->cpu_bstats,
 404                                         &p->tcfa_rate_est,
 405                                         &p->tcfa_lock, NULL, est);
 406                 if (err)
 407                         goto err3;
 408         }
 409
 410         p->idrinfo = idrinfo;
 411         p->ops = ops;
 412         INIT_LIST_HEAD(&p->list);
 413         *a = p;
 414         return 0;
 415 err3:
 416         free_percpu(p->cpu_qstats);
 417 err2:
 418         free_percpu(p->cpu_bstats);
 419 err1:
 420         kfree(p);
 421         return err;
 422 }
 423 EXPORT_SYMBOL(tcf_idr_create);
 424
 425 void tcf_idr_insert(struct tc_action_net *tn, struct tc_action *a)
 426 {
 427         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 428
 429         spin_lock(&idrinfo->lock);
 430         /* Replace ERR_PTR(-EBUSY) allocated by tcf_idr_check_alloc */
 431         WARN_ON(!IS_ERR(idr_replace(&idrinfo->action_idr, a, a->tcfa_index)));
 432         spin_unlock(&idrinfo->lock);
 433 }
 434 EXPORT_SYMBOL(tcf_idr_insert);
 435
 436 /* Cleanup idr index that was allocated but not initialized. */
 437
 438 void tcf_idr_cleanup(struct tc_action_net *tn, u32 index)
 439 {
 440         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 441
 442         spin_lock(&idrinfo->lock);
 443         /* Remove ERR_PTR(-EBUSY) allocated by tcf_idr_check_alloc */
 444         WARN_ON(!IS_ERR(idr_remove(&idrinfo->action_idr, index)));
 445         spin_unlock(&idrinfo->lock);
 446 }
 447 EXPORT_SYMBOL(tcf_idr_cleanup);
 448
 449 /* Check if action with specified index exists. If actions is found, increments
 450  * its reference and bind counters, and return 1. Otherwise insert temporary
 451  * error pointer (to prevent concurrent users from inserting actions with same
 452  * index) and return 0.
 453  */
 454
 455 int tcf_idr_check_alloc(struct tc_action_net *tn, u32 *index,
 456                         struct tc_action **a, int bind)
 457 {
 458         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 459         struct tc_action *p;
 460         int ret;
 461
 462 again:
 463         spin_lock(&idrinfo->lock);
 464         if (*index) {
 465                 p = idr_find(&idrinfo->action_idr, *index);
 466                 if (IS_ERR(p)) {
 467                         /* This means that another process allocated
 468                          * index but did not assign the pointer yet.
 469                          */
 470                         spin_unlock(&idrinfo->lock);
 471                         goto again;
 472                 }
 473
 474                 if (p) {
 475                         refcount_inc(&p->tcfa_refcnt);
 476                         if (bind)
 477                                 atomic_inc(&p->tcfa_bindcnt);
 478                         *a = p;
 479                         ret = 1;
 480                 } else {
 481                         *a = NULL;
 482                         ret = idr_alloc_u32(&idrinfo->action_idr, NULL, index,
 483                                             *index, GFP_ATOMIC);
 484                         if (!ret)
 485                                 idr_replace(&idrinfo->action_idr,
 486                                             ERR_PTR(-EBUSY), *index);
 487                 }
 488         } else {
 489                 *index = 1;
 490                 *a = NULL;
 491                 ret = idr_alloc_u32(&idrinfo->action_idr, NULL, index,
 492                                     UINT_MAX, GFP_ATOMIC);
 493                 if (!ret)
 494                         idr_replace(&idrinfo->action_idr, ERR_PTR(-EBUSY),
 495                                     *index);
 496         }
 497         spin_unlock(&idrinfo->lock);
 498         return ret;
 499 }
 500 EXPORT_SYMBOL(tcf_idr_check_alloc);
 501
 502 void tcf_idrinfo_destroy(const struct tc_action_ops *ops,
 503                          struct tcf_idrinfo *idrinfo)
 504 {
 505         struct idr *idr = &idrinfo->action_idr;
 506         struct tc_action *p;
 507         int ret;
 508         unsigned long id = 1;
 509
 510         idr_for_each_entry_ul(idr, p, id) {
 511                 ret = __tcf_idr_release(p, false, true);
 512                 if (ret == ACT_P_DELETED)
 513                         module_put(ops->owner);
 514                 else if (ret < 0)
 515                         return;
 516         }
 517         idr_destroy(&idrinfo->action_idr);
 518 }
 519 EXPORT_SYMBOL(tcf_idrinfo_destroy);
 520
 521 static LIST_HEAD(act_base);
 522 static DEFINE_RWLOCK(act_mod_lock);
 523
 524 int tcf_register_action(struct tc_action_ops *act,
 525                         struct pernet_operations *ops)
 526 {
 527         struct tc_action_ops *a;
 528         int ret;
 529
 530         if (!act->act || !act->dump || !act->init || !act->walk || !act->lookup)
 531                 return -EINVAL;
 532
 533         /* We have to register pernet ops before making the action ops visible,
 534          * otherwise tcf_action_init_1() could get a partially initialized
 535          * netns.
 536          */
 537         ret = register_pernet_subsys(ops);
 538         if (ret)
 539                 return ret;
 540
 541         write_lock(&act_mod_lock);
 542         list_for_each_entry(a, &act_base, head) {
 543                 if (act->type == a->type || (strcmp(act->kind, a->kind) == 0)) {
 544                         write_unlock(&act_mod_lock);
 545                         unregister_pernet_subsys(ops);
 546                         return -EEXIST;
 547                 }
 548         }
 549         list_add_tail(&act->head, &act_base);
 550         write_unlock(&act_mod_lock);
 551
 552         return 0;
 553 }
 554 EXPORT_SYMBOL(tcf_register_action);
 555
 556 int tcf_unregister_action(struct tc_action_ops *act,
 557                           struct pernet_operations *ops)
 558 {
 559         struct tc_action_ops *a;
 560         int err = -ENOENT;
 561
 562         write_lock(&act_mod_lock);
 563         list_for_each_entry(a, &act_base, head) {
 564                 if (a == act) {
 565                         list_del(&act->head);
 566                         err = 0;
 567                         break;
 568                 }
 569         }
 570         write_unlock(&act_mod_lock);
 571         if (!err)
 572                 unregister_pernet_subsys(ops);
 573         return err;
 574 }
 575 EXPORT_SYMBOL(tcf_unregister_action);
 576
 577 /* lookup by name */
 578 static struct tc_action_ops *tc_lookup_action_n(char *kind)
 579 {
 580         struct tc_action_ops *a, *res = NULL;
 581
 582         if (kind) {
 583                 read_lock(&act_mod_lock);
 584                 list_for_each_entry(a, &act_base, head) {
 585                         if (strcmp(kind, a->kind) == 0) {
 586                                 if (try_module_get(a->owner))
 587                                         res = a;
 588                                 break;
 589                         }
 590                 }
 591                 read_unlock(&act_mod_lock);
 592         }
 593         return res;
 594 }
 595
 596 /* lookup by nlattr */
 597 static struct tc_action_ops *tc_lookup_action(struct nlattr *kind)
 598 {
 599         struct tc_action_ops *a, *res = NULL;
 600
 601         if (kind) {
 602                 read_lock(&act_mod_lock);
 603                 list_for_each_entry(a, &act_base, head) {
 604                         if (nla_strcmp(kind, a->kind) == 0) {
 605                                 if (try_module_get(a->owner))
 606                                         res = a;
 607                                 break;
 608                         }
 609                 }
 610                 read_unlock(&act_mod_lock);
 611         }
 612         return res;
 613 }
 614
 615 /*TCA_ACT_MAX_PRIO is 32, there count upto 32 */
 616 #define TCA_ACT_MAX_PRIO_MASK 0x1FF
 617 int tcf_action_exec(struct sk_buff *skb, struct tc_action **actions,
 618                     int nr_actions, struct tcf_result *res)
 619 {
 620         u32 jmp_prgcnt = 0;
 621         u32 jmp_ttl = TCA_ACT_MAX_PRIO; /*matches actions per filter */
 622         int i;
 623         int ret = TC_ACT_OK;
 624
 625         if (skb_skip_tc_classify(skb))
 626                 return TC_ACT_OK;
 627
 628 restart_act_graph:
 629         for (i = 0; i < nr_actions; i++) {
 630                 const struct tc_action *a = actions[i];
 631
 632                 if (jmp_prgcnt > 0) {
 633                         jmp_prgcnt -= 1;
 634                         continue;
 635                 }
 636 repeat:
 637                 ret = a->ops->act(skb, a, res);
 638                 if (ret == TC_ACT_REPEAT)
 639                         goto repeat;    /* we need a ttl - JHS */
 640
 641                 if (TC_ACT_EXT_CMP(ret, TC_ACT_JUMP)) {
 642                         jmp_prgcnt = ret & TCA_ACT_MAX_PRIO_MASK;
 643                         if (!jmp_prgcnt || (jmp_prgcnt > nr_actions)) {
 644                                 /* faulty opcode, stop pipeline */
 645                                 return TC_ACT_OK;
 646                         } else {
 647                                 jmp_ttl -= 1;
 648                                 if (jmp_ttl > 0)
 649                                         goto restart_act_graph;
 650                                 else /* faulty graph, stop pipeline */
 651                                         return TC_ACT_OK;
 652                         }
 653                 } else if (TC_ACT_EXT_CMP(ret, TC_ACT_GOTO_CHAIN)) {
 654                         tcf_action_goto_chain_exec(a, res);
 655                 }
 656
 657                 if (ret != TC_ACT_PIPE)
 658                         break;
 659         }
 660
 661         return ret;
 662 }
 663 EXPORT_SYMBOL(tcf_action_exec);
 664
 665 int tcf_action_destroy(struct tc_action *actions[], int bind)
 666 {
 667         const struct tc_action_ops *ops;
 668         struct tc_action *a;
 669         int ret = 0, i;
 670
 671         for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
 672                 a = actions[i];
 673                 actions[i] = NULL;
 674                 ops = a->ops;
 675                 ret = __tcf_idr_release(a, bind, true);
 676                 if (ret == ACT_P_DELETED)
 677                         module_put(ops->owner);
 678                 else if (ret < 0)
 679                         return ret;
 680         }
 681         return ret;
 682 }
 683
 684 static int tcf_action_put(struct tc_action *p)
 685 {
 686         return __tcf_action_put(p, false);
 687 }
 688
 689 static void tcf_action_put_many(struct tc_action *actions[])
 690 {
 691         int i;
 692
 693         for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
 694                 struct tc_action *a = actions[i];
 695                 const struct tc_action_ops *ops = a->ops;
 696
 697                 if (tcf_action_put(a))
 698                         module_put(ops->owner);
 699         }
 700 }
 701
 702 int
 703 tcf_action_dump_old(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
 704 {
 705         return a->ops->dump(skb, a, bind, ref);
 706 }
 707
 708 int
 709 tcf_action_dump_1(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
 710 {
 711         int err = -EINVAL;
 712         unsigned char *b = skb_tail_pointer(skb);
 713         struct nlattr *nest;
 714         struct tc_cookie *cookie;
 715
 716         if (nla_put_string(skb, TCA_KIND, a->ops->kind))
 717                 goto nla_put_failure;
 718         if (tcf_action_copy_stats(skb, a, 0))
 719                 goto nla_put_failure;
 720
 721         rcu_read_lock();
 722         cookie = rcu_dereference(a->act_cookie);
 723         if (cookie) {
 724                 if (nla_put(skb, TCA_ACT_COOKIE, cookie->len, cookie->data)) {
 725                         rcu_read_unlock();
 726                         goto nla_put_failure;
 727                 }
 728         }
 729         rcu_read_unlock();
 730
 731         nest = nla_nest_start(skb, TCA_OPTIONS);
 732         if (nest == NULL)
 733                 goto nla_put_failure;
 734         err = tcf_action_dump_old(skb, a, bind, ref);
 735         if (err > 0) {
 736                 nla_nest_end(skb, nest);
 737                 return err;
 738         }
 739
 740 nla_put_failure:
 741         nlmsg_trim(skb, b);
 742         return -1;
 743 }
 744 EXPORT_SYMBOL(tcf_action_dump_1);
 745
 746 int tcf_action_dump(struct sk_buff *skb, struct tc_action *actions[],
 747                     int bind, int ref)
 748 {
 749         struct tc_action *a;
 750         int err = -EINVAL, i;
 751         struct nlattr *nest;
 752
 753         for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
 754                 a = actions[i];
 755                 nest = nla_nest_start(skb, a->order);
 756                 if (nest == NULL)
 757                         goto nla_put_failure;
 758                 err = tcf_action_dump_1(skb, a, bind, ref);
 759                 if (err < 0)
 760                         goto errout;
 761                 nla_nest_end(skb, nest);
 762         }
 763
 764         return 0;
 765
 766 nla_put_failure:
 767         err = -EINVAL;
 768 errout:
 769         nla_nest_cancel(skb, nest);
 770         return err;
 771 }
 772
 773 static struct tc_cookie *nla_memdup_cookie(struct nlattr **tb)
 774 {
 775         struct tc_cookie *c = kzalloc(sizeof(*c), GFP_KERNEL);
 776         if (!c)
 777                 return NULL;
 778
 779         c->data = nla_memdup(tb[TCA_ACT_COOKIE], GFP_KERNEL);
 780         if (!c->data) {
 781                 kfree(c);
 782                 return NULL;
 783         }
 784         c->len = nla_len(tb[TCA_ACT_COOKIE]);
 785
 786         return c;
 787 }
 788
 789 static bool tcf_action_valid(int action)
 790 {
 791         int opcode = TC_ACT_EXT_OPCODE(action);
 792
 793         if (!opcode)
 794                 return action <= TC_ACT_VALUE_MAX;
 795         return opcode <= TC_ACT_EXT_OPCODE_MAX || action == TC_ACT_UNSPEC;
 796 }
 797
 798 struct tc_action *tcf_action_init_1(struct net *net, struct tcf_proto *tp,
 799                                     struct nlattr *nla, struct nlattr *est,
 800                                     char *name, int ovr, int bind,
 801                                     bool rtnl_held,
 802                                     struct netlink_ext_ack *extack)
 803 {
 804         struct tc_action *a;
 805         struct tc_action_ops *a_o;
 806         struct tc_cookie *cookie = NULL;
 807         char act_name[IFNAMSIZ];
 808         struct nlattr *tb[TCA_ACT_MAX + 1];
 809         struct nlattr *kind;
 810         int err;
 811
 812         if (name == NULL) {
 813                 err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, extack);
 814                 if (err < 0)
 815                         goto err_out;
 816                 err = -EINVAL;
 817                 kind = tb[TCA_ACT_KIND];
 818                 if (!kind) {
 819                         NL_SET_ERR_MSG(extack, "TC action kind must be specified");
 820                         goto err_out;
 821                 }
 822                 if (nla_strlcpy(act_name, kind, IFNAMSIZ) >= IFNAMSIZ) {
 823                         NL_SET_ERR_MSG(extack, "TC action name too long");
 824                         goto err_out;
 825                 }
 826                 if (tb[TCA_ACT_COOKIE]) {
 827                         int cklen = nla_len(tb[TCA_ACT_COOKIE]);
 828
 829                         if (cklen > TC_COOKIE_MAX_SIZE) {
 830                                 NL_SET_ERR_MSG(extack, "TC cookie size above the maximum");
 831                                 goto err_out;
 832                         }
 833
 834                         cookie = nla_memdup_cookie(tb);
 835                         if (!cookie) {
 836                                 NL_SET_ERR_MSG(extack, "No memory to generate TC cookie");
 837                                 err = -ENOMEM;
 838                                 goto err_out;
 839                         }
 840                 }
 841         } else {
 842                 if (strlcpy(act_name, name, IFNAMSIZ) >= IFNAMSIZ) {
 843                         NL_SET_ERR_MSG(extack, "TC action name too long");
 844                         err = -EINVAL;
 845                         goto err_out;
 846                 }
 847         }
 848
 849         a_o = tc_lookup_action_n(act_name);
 850         if (a_o == NULL) {
 851 #ifdef CONFIG_MODULES
 852                 if (rtnl_held)
 853                         rtnl_unlock();
 854                 request_module("act_%s", act_name);
 855                 if (rtnl_held)
 856                         rtnl_lock();
 857
 858                 a_o = tc_lookup_action_n(act_name);
 859
 860                 /* We dropped the RTNL semaphore in order to
 861                  * perform the module load.  So, even if we
 862                  * succeeded in loading the module we have to
 863                  * tell the caller to replay the request.  We
 864                  * indicate this using -EAGAIN.
 865                  */
 866                 if (a_o != NULL) {
 867                         err = -EAGAIN;
 868                         goto err_mod;
 869                 }
 870 #endif
 871                 NL_SET_ERR_MSG(extack, "Failed to load TC action module");
 872                 err = -ENOENT;
 873                 goto err_out;
 874         }
 875
 876         /* backward compatibility for policer */
 877         if (name == NULL)
 878                 err = a_o->init(net, tb[TCA_ACT_OPTIONS], est, &a, ovr, bind,
 879                                 rtnl_held, extack);
 880         else
 881                 err = a_o->init(net, nla, est, &a, ovr, bind, rtnl_held,
 882                                 extack);
 883         if (err < 0)
 884                 goto err_mod;
 885
 886         if (!name && tb[TCA_ACT_COOKIE])
 887                 tcf_set_action_cookie(&a->act_cookie, cookie);
 888
 889         /* module count goes up only when brand new policy is created
 890          * if it exists and is only bound to in a_o->init() then
 891          * ACT_P_CREATED is not returned (a zero is).
 892          */
 893         if (err != ACT_P_CREATED)
 894                 module_put(a_o->owner);
 895
 896         if (TC_ACT_EXT_CMP(a->tcfa_action, TC_ACT_GOTO_CHAIN)) {
 897                 err = tcf_action_goto_chain_init(a, tp);
 898                 if (err) {
 899                         struct tc_action *actions[] = { a, NULL };
 900
 901                         tcf_action_destroy(actions, bind);
 902                         NL_SET_ERR_MSG(extack, "Failed to init TC action chain");
 903                         return ERR_PTR(err);
 904                 }
 905         }
 906
 907         if (!tcf_action_valid(a->tcfa_action)) {
 908                 NL_SET_ERR_MSG(extack, "invalid action value, using TC_ACT_UNSPEC instead");
 909                 a->tcfa_action = TC_ACT_UNSPEC;
 910         }
 911
 912         return a;
 913
 914 err_mod:
 915         module_put(a_o->owner);
 916 err_out:
 917         if (cookie) {
 918                 kfree(cookie->data);
 919                 kfree(cookie);
 920         }
 921         return ERR_PTR(err);
 922 }
 923
 924 /* Returns numbers of initialized actions or negative error. */
 925
 926 int tcf_action_init(struct net *net, struct tcf_proto *tp, struct nlattr *nla,
 927                     struct nlattr *est, char *name, int ovr, int bind,
 928                     struct tc_action *actions[], size_t *attr_size,
 929                     bool rtnl_held, struct netlink_ext_ack *extack)
 930 {
 931         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
 932         struct tc_action *act;
 933         size_t sz = 0;
 934         int err;
 935         int i;
 936
 937         err = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL, extack);
 938         if (err < 0)
 939                 return err;
 940
 941         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
 942                 act = tcf_action_init_1(net, tp, tb[i], est, name, ovr, bind,
 943                                         rtnl_held, extack);
 944                 if (IS_ERR(act)) {
 945                         err = PTR_ERR(act);
 946                         goto err;
 947                 }
 948                 act->order = i;
 949                 sz += tcf_action_fill_size(act);
 950                 /* Start from index 0 */
 951                 actions[i - 1] = act;
 952         }
 953
 954         *attr_size = tcf_action_full_attrs_size(sz);
 955         return i - 1;
 956
 957 err:
 958         tcf_action_destroy(actions, bind);
 959         return err;
 960 }
 961
 962 int tcf_action_copy_stats(struct sk_buff *skb, struct tc_action *p,
 963                           int compat_mode)
 964 {
 965         int err = 0;
 966         struct gnet_dump d;
 967
 968         if (p == NULL)
 969                 goto errout;
 970
 971         /* compat_mode being true specifies a call that is supposed
 972          * to add additional backward compatibility statistic TLVs.
 973          */
 974         if (compat_mode) {
 975                 if (p->type == TCA_OLD_COMPAT)
 976                         err = gnet_stats_start_copy_compat(skb, 0,
 977                                                            TCA_STATS,
 978                                                            TCA_XSTATS,
 979                                                            &p->tcfa_lock, &d,
 980                                                            TCA_PAD);
 981                 else
 982                         return 0;
 983         } else
 984                 err = gnet_stats_start_copy(skb, TCA_ACT_STATS,
 985                                             &p->tcfa_lock, &d, TCA_ACT_PAD);
 986
 987         if (err < 0)
 988                 goto errout;
 989
 990         if (gnet_stats_copy_basic(NULL, &d, p->cpu_bstats, &p->tcfa_bstats) < 0 ||
 991             gnet_stats_copy_rate_est(&d, &p->tcfa_rate_est) < 0 ||
 992             gnet_stats_copy_queue(&d, p->cpu_qstats,
 993                                   &p->tcfa_qstats,
 994                                   p->tcfa_qstats.qlen) < 0)
 995                 goto errout;
 996
 997         if (gnet_stats_finish_copy(&d) < 0)
 998                 goto errout;
 999
1000         return 0;
1001
1002 errout:
1003         return -1;
1004 }
1005
1006 static int tca_get_fill(struct sk_buff *skb, struct tc_action *actions[],
1007                         u32 portid, u32 seq, u16 flags, int event, int bind,
1008                         int ref)
1009 {
1010         struct tcamsg *t;
1011         struct nlmsghdr *nlh;
1012         unsigned char *b = skb_tail_pointer(skb);
1013         struct nlattr *nest;
1014
1015         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*t), flags);
1016         if (!nlh)
1017                 goto out_nlmsg_trim;
1018         t = nlmsg_data(nlh);
1019         t->tca_family = AF_UNSPEC;
1020         t->tca__pad1 = 0;
1021         t->tca__pad2 = 0;
1022
1023         nest = nla_nest_start(skb, TCA_ACT_TAB);
1024         if (!nest)
1025                 goto out_nlmsg_trim;
1026
1027         if (tcf_action_dump(skb, actions, bind, ref) < 0)
1028                 goto out_nlmsg_trim;
1029
1030         nla_nest_end(skb, nest);
1031
1032         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1033         return skb->len;
1034
1035 out_nlmsg_trim:
1036         nlmsg_trim(skb, b);
1037         return -1;
1038 }
1039
1040 static int
1041 tcf_get_notify(struct net *net, u32 portid, struct nlmsghdr *n,
1042                struct tc_action *actions[], int event,
1043                struct netlink_ext_ack *extack)
1044 {
1045         struct sk_buff *skb;
1046
1047         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1048         if (!skb)
1049                 return -ENOBUFS;
1050         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, event,
1051                          0, 1) <= 0) {
1052                 NL_SET_ERR_MSG(extack, "Failed to fill netlink attributes while adding TC action");
1053                 kfree_skb(skb);
1054                 return -EINVAL;
1055         }
1056
1057         return rtnl_unicast(skb, net, portid);
1058 }
1059
1060 static struct tc_action *tcf_action_get_1(struct net *net, struct nlattr *nla,
1061                                           struct nlmsghdr *n, u32 portid,
1062                                           struct netlink_ext_ack *extack)
1063 {
1064         struct nlattr *tb[TCA_ACT_MAX + 1];
1065         const struct tc_action_ops *ops;
1066         struct tc_action *a;
1067         int index;
1068         int err;
1069
1070         err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, extack);
1071         if (err < 0)
1072                 goto err_out;
1073
1074         err = -EINVAL;
1075         if (tb[TCA_ACT_INDEX] == NULL ||
1076             nla_len(tb[TCA_ACT_INDEX]) < sizeof(index)) {
1077                 NL_SET_ERR_MSG(extack, "Invalid TC action index value");
1078                 goto err_out;
1079         }
1080         index = nla_get_u32(tb[TCA_ACT_INDEX]);
1081
1082         err = -EINVAL;
1083         ops = tc_lookup_action(tb[TCA_ACT_KIND]);
1084         if (!ops) { /* could happen in batch of actions */
1085                 NL_SET_ERR_MSG(extack, "Specified TC action not found");
1086                 goto err_out;
1087         }
1088         err = -ENOENT;
1089         if (ops->lookup(net, &a, index, extack) == 0)
1090                 goto err_mod;
1091
1092         module_put(ops->owner);
1093         return a;
1094
1095 err_mod:
1096         module_put(ops->owner);
1097 err_out:
1098         return ERR_PTR(err);
1099 }
1100
1101 static int tca_action_flush(struct net *net, struct nlattr *nla,
1102                             struct nlmsghdr *n, u32 portid,
1103                             struct netlink_ext_ack *extack)
1104 {
1105         struct sk_buff *skb;
1106         unsigned char *b;
1107         struct nlmsghdr *nlh;
1108         struct tcamsg *t;
1109         struct netlink_callback dcb;
1110         struct nlattr *nest;
1111         struct nlattr *tb[TCA_ACT_MAX + 1];
1112         const struct tc_action_ops *ops;
1113         struct nlattr *kind;
1114         int err = -ENOMEM;
1115
1116         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1117         if (!skb)
1118                 return err;
1119
1120         b = skb_tail_pointer(skb);
1121
1122         err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, extack);
1123         if (err < 0)
1124                 goto err_out;
1125
1126         err = -EINVAL;
1127         kind = tb[TCA_ACT_KIND];
1128         ops = tc_lookup_action(kind);
1129         if (!ops) { /*some idjot trying to flush unknown action */
1130                 NL_SET_ERR_MSG(extack, "Cannot flush unknown TC action");
1131                 goto err_out;
1132         }
1133
1134         nlh = nlmsg_put(skb, portid, n->nlmsg_seq, RTM_DELACTION,
1135                         sizeof(*t), 0);
1136         if (!nlh) {
1137                 NL_SET_ERR_MSG(extack, "Failed to create TC action flush notification");
1138                 goto out_module_put;
1139         }
1140         t = nlmsg_data(nlh);
1141         t->tca_family = AF_UNSPEC;
1142         t->tca__pad1 = 0;
1143         t->tca__pad2 = 0;
1144
1145         nest = nla_nest_start(skb, TCA_ACT_TAB);
1146         if (!nest) {
1147                 NL_SET_ERR_MSG(extack, "Failed to add new netlink message");
1148                 goto out_module_put;
1149         }
1150
1151         err = ops->walk(net, skb, &dcb, RTM_DELACTION, ops, extack);
1152         if (err <= 0) {
1153                 nla_nest_cancel(skb, nest);
1154                 goto out_module_put;
1155         }
1156
1157         nla_nest_end(skb, nest);
1158
1159         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1160         nlh->nlmsg_flags |= NLM_F_ROOT;
1161         module_put(ops->owner);
1162         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1163                              n->nlmsg_flags & NLM_F_ECHO);
1164         if (err > 0)
1165                 return 0;
1166         if (err < 0)
1167                 NL_SET_ERR_MSG(extack, "Failed to send TC action flush notification");
1168
1169         return err;
1170
1171 out_module_put:
1172         module_put(ops->owner);
1173 err_out:
1174         kfree_skb(skb);
1175         return err;
1176 }
1177
1178 static int tcf_action_delete(struct net *net, struct tc_action *actions[],
1179                              int *acts_deleted, struct netlink_ext_ack *extack)
1180 {
1181         u32 act_index;
1182         int ret, i;
1183
1184         for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
1185                 struct tc_action *a = actions[i];
1186                 const struct tc_action_ops *ops = a->ops;
1187
1188                 /* Actions can be deleted concurrently so we must save their
1189                  * type and id to search again after reference is released.
1190                  */
1191                 act_index = a->tcfa_index;
1192
1193                 if (tcf_action_put(a)) {
1194                         /* last reference, action was deleted concurrently */
1195                         module_put(ops->owner);
1196                 } else  {
1197                         /* now do the delete */
1198                         ret = ops->delete(net, act_index);
1199                         if (ret < 0) {
1200                                 *acts_deleted = i + 1;
1201                                 return ret;
1202                         }
1203                 }
1204         }
1205         *acts_deleted = i;
1206         return 0;
1207 }
1208
1209 static int
1210 tcf_del_notify(struct net *net, struct nlmsghdr *n, struct tc_action *actions[],
1211                int *acts_deleted, u32 portid, size_t attr_size,
1212                struct netlink_ext_ack *extack)
1213 {
1214         int ret;
1215         struct sk_buff *skb;
1216
1217         skb = alloc_skb(attr_size <= NLMSG_GOODSIZE ? NLMSG_GOODSIZE : attr_size,
1218                         GFP_KERNEL);
1219         if (!skb)
1220                 return -ENOBUFS;
1221
1222         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, RTM_DELACTION,
1223                          0, 2) <= 0) {
1224                 NL_SET_ERR_MSG(extack, "Failed to fill netlink TC action attributes");
1225                 kfree_skb(skb);
1226                 return -EINVAL;
1227         }
1228
1229         /* now do the delete */
1230         ret = tcf_action_delete(net, actions, acts_deleted, extack);
1231         if (ret < 0) {
1232                 NL_SET_ERR_MSG(extack, "Failed to delete TC action");
1233                 kfree_skb(skb);
1234                 return ret;
1235         }
1236
1237         ret = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1238                              n->nlmsg_flags & NLM_F_ECHO);
1239         if (ret > 0)
1240                 return 0;
1241         return ret;
1242 }
1243
1244 static int
1245 tca_action_gd(struct net *net, struct nlattr *nla, struct nlmsghdr *n,
1246               u32 portid, int event, struct netlink_ext_ack *extack)
1247 {
1248         int i, ret;
1249         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1250         struct tc_action *act;
1251         size_t attr_size = 0;
1252         struct tc_action *actions[TCA_ACT_MAX_PRIO + 1] = {};
1253         int acts_deleted = 0;
1254
1255         ret = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL, extack);
1256         if (ret < 0)
1257                 return ret;
1258
1259         if (event == RTM_DELACTION && n->nlmsg_flags & NLM_F_ROOT) {
1260                 if (tb[1])
1261                         return tca_action_flush(net, tb[1], n, portid, extack);
1262
1263                 NL_SET_ERR_MSG(extack, "Invalid netlink attributes while flushing TC action");
1264                 return -EINVAL;
1265         }
1266
1267         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
1268                 act = tcf_action_get_1(net, tb[i], n, portid, extack);
1269                 if (IS_ERR(act)) {
1270                         ret = PTR_ERR(act);
1271                         goto err;
1272                 }
1273                 act->order = i;
1274                 attr_size += tcf_action_fill_size(act);
1275                 actions[i - 1] = act;
1276         }
1277
1278         attr_size = tcf_action_full_attrs_size(attr_size);
1279
1280         if (event == RTM_GETACTION)
1281                 ret = tcf_get_notify(net, portid, n, actions, event, extack);
1282         else { /* delete */
1283                 ret = tcf_del_notify(net, n, actions, &acts_deleted, portid,
1284                                      attr_size, extack);
1285                 if (ret)
1286                         goto err;
1287                 return ret;
1288         }
1289 err:
1290         tcf_action_put_many(&actions[acts_deleted]);
1291         return ret;
1292 }
1293
1294 static int
1295 tcf_add_notify(struct net *net, struct nlmsghdr *n, struct tc_action *actions[],
1296                u32 portid, size_t attr_size, struct netlink_ext_ack *extack)
1297 {
1298         struct sk_buff *skb;
1299         int err = 0;
1300
1301         skb = alloc_skb(attr_size <= NLMSG_GOODSIZE ? NLMSG_GOODSIZE : attr_size,
1302                         GFP_KERNEL);
1303         if (!skb)
1304                 return -ENOBUFS;
1305
1306         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, n->nlmsg_flags,
1307                          RTM_NEWACTION, 0, 0) <= 0) {
1308                 NL_SET_ERR_MSG(extack, "Failed to fill netlink attributes while adding TC action");
1309                 kfree_skb(skb);
1310                 return -EINVAL;
1311         }
1312
1313         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1314                              n->nlmsg_flags & NLM_F_ECHO);
1315         if (err > 0)
1316                 err = 0;
1317         return err;
1318 }
1319
1320 static int tcf_action_add(struct net *net, struct nlattr *nla,
1321                           struct nlmsghdr *n, u32 portid, int ovr,
1322                           struct netlink_ext_ack *extack)
1323 {
1324         size_t attr_size = 0;
1325         int ret = 0;
1326         struct tc_action *actions[TCA_ACT_MAX_PRIO] = {};
1327
1328         ret = tcf_action_init(net, NULL, nla, NULL, NULL, ovr, 0, actions,
1329                               &attr_size, true, extack);
1330         if (ret < 0)
1331                 return ret;
1332         ret = tcf_add_notify(net, n, actions, portid, attr_size, extack);
1333         if (ovr)
1334                 tcf_action_put_many(actions);
1335
1336         return ret;
1337 }
1338
1339 static u32 tcaa_root_flags_allowed = TCA_FLAG_LARGE_DUMP_ON;
1340 static const struct nla_policy tcaa_policy[TCA_ROOT_MAX + 1] = {
1341         [TCA_ROOT_FLAGS] = { .type = NLA_BITFIELD32,
1342                              .validation_data = &tcaa_root_flags_allowed },
1343         [TCA_ROOT_TIME_DELTA]      = { .type = NLA_U32 },
1344 };
1345
1346 static int tc_ctl_action(struct sk_buff *skb, struct nlmsghdr *n,
1347                          struct netlink_ext_ack *extack)
1348 {
1349         struct net *net = sock_net(skb->sk);
1350         struct nlattr *tca[TCA_ROOT_MAX + 1];
1351         u32 portid = skb ? NETLINK_CB(skb).portid : 0;
1352         int ret = 0, ovr = 0;
1353
1354         if ((n->nlmsg_type != RTM_GETACTION) &&
1355             !netlink_capable(skb, CAP_NET_ADMIN))
1356                 return -EPERM;
1357
1358         ret = nlmsg_parse(n, sizeof(struct tcamsg), tca, TCA_ROOT_MAX, NULL,
1359                           extack);
1360         if (ret < 0)
1361                 return ret;
1362
1363         if (tca[TCA_ACT_TAB] == NULL) {
1364                 NL_SET_ERR_MSG(extack, "Netlink action attributes missing");
1365                 return -EINVAL;
1366         }
1367
1368         /* n->nlmsg_flags & NLM_F_CREATE */
1369         switch (n->nlmsg_type) {
1370         case RTM_NEWACTION:
1371                 /* we are going to assume all other flags
1372                  * imply create only if it doesn't exist
1373                  * Note that CREATE | EXCL implies that
1374                  * but since we want avoid ambiguity (eg when flags
1375                  * is zero) then just set this
1376                  */
1377                 if (n->nlmsg_flags & NLM_F_REPLACE)
1378                         ovr = 1;
1379 replay:
1380                 ret = tcf_action_add(net, tca[TCA_ACT_TAB], n, portid, ovr,
1381                                      extack);
1382                 if (ret == -EAGAIN)
1383                         goto replay;
1384                 break;
1385         case RTM_DELACTION:
1386                 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1387                                     portid, RTM_DELACTION, extack);
1388                 break;
1389         case RTM_GETACTION:
1390                 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1391                                     portid, RTM_GETACTION, extack);
1392                 break;
1393         default:
1394                 BUG();
1395         }
1396
1397         return ret;
1398 }
1399
1400 static struct nlattr *find_dump_kind(struct nlattr **nla)
1401 {
1402         struct nlattr *tb1, *tb2[TCA_ACT_MAX + 1];
1403         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1404         struct nlattr *kind;
1405
1406         tb1 = nla[TCA_ACT_TAB];
1407         if (tb1 == NULL)
1408                 return NULL;
1409
1410         if (nla_parse(tb, TCA_ACT_MAX_PRIO, nla_data(tb1),
1411                       NLMSG_ALIGN(nla_len(tb1)), NULL, NULL) < 0)
1412                 return NULL;
1413
1414         if (tb[1] == NULL)
1415                 return NULL;
1416         if (nla_parse_nested(tb2, TCA_ACT_MAX, tb[1], NULL, NULL) < 0)
1417                 return NULL;
1418         kind = tb2[TCA_ACT_KIND];
1419
1420         return kind;
1421 }
1422
1423 static int tc_dump_action(struct sk_buff *skb, struct netlink_callback *cb)
1424 {
1425         struct net *net = sock_net(skb->sk);
1426         struct nlmsghdr *nlh;
1427         unsigned char *b = skb_tail_pointer(skb);
1428         struct nlattr *nest;
1429         struct tc_action_ops *a_o;
1430         int ret = 0;
1431         struct tcamsg *t = (struct tcamsg *) nlmsg_data(cb->nlh);
1432         struct nlattr *tb[TCA_ROOT_MAX + 1];
1433         struct nlattr *count_attr = NULL;
1434         unsigned long jiffy_since = 0;
1435         struct nlattr *kind = NULL;
1436         struct nla_bitfield32 bf;
1437         u32 msecs_since = 0;
1438         u32 act_count = 0;
1439
1440         ret = nlmsg_parse(cb->nlh, sizeof(struct tcamsg), tb, TCA_ROOT_MAX,
1441                           tcaa_policy, NULL);
1442         if (ret < 0)
1443                 return ret;
1444
1445         kind = find_dump_kind(tb);
1446         if (kind == NULL) {
1447                 pr_info("tc_dump_action: action bad kind\n");
1448                 return 0;
1449         }
1450
1451         a_o = tc_lookup_action(kind);
1452         if (a_o == NULL)
1453                 return 0;
1454
1455         cb->args[2] = 0;
1456         if (tb[TCA_ROOT_FLAGS]) {
1457                 bf = nla_get_bitfield32(tb[TCA_ROOT_FLAGS]);
1458                 cb->args[2] = bf.value;
1459         }
1460
1461         if (tb[TCA_ROOT_TIME_DELTA]) {
1462                 msecs_since = nla_get_u32(tb[TCA_ROOT_TIME_DELTA]);
1463         }
1464
1465         nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1466                         cb->nlh->nlmsg_type, sizeof(*t), 0);
1467         if (!nlh)
1468                 goto out_module_put;
1469
1470         if (msecs_since)
1471                 jiffy_since = jiffies - msecs_to_jiffies(msecs_since);
1472
1473         t = nlmsg_data(nlh);
1474         t->tca_family = AF_UNSPEC;
1475         t->tca__pad1 = 0;
1476         t->tca__pad2 = 0;
1477         cb->args[3] = jiffy_since;
1478         count_attr = nla_reserve(skb, TCA_ROOT_COUNT, sizeof(u32));
1479         if (!count_attr)
1480                 goto out_module_put;
1481
1482         nest = nla_nest_start(skb, TCA_ACT_TAB);
1483         if (nest == NULL)
1484                 goto out_module_put;
1485
1486         ret = a_o->walk(net, skb, cb, RTM_GETACTION, a_o, NULL);
1487         if (ret < 0)
1488                 goto out_module_put;
1489
1490         if (ret > 0) {
1491                 nla_nest_end(skb, nest);
1492                 ret = skb->len;
1493                 act_count = cb->args[1];
1494                 memcpy(nla_data(count_attr), &act_count, sizeof(u32));
1495                 cb->args[1] = 0;
1496         } else
1497                 nlmsg_trim(skb, b);
1498
1499         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1500         if (NETLINK_CB(cb->skb).portid && ret)
1501                 nlh->nlmsg_flags |= NLM_F_MULTI;
1502         module_put(a_o->owner);
1503         return skb->len;
1504
1505 out_module_put:
1506         module_put(a_o->owner);
1507         nlmsg_trim(skb, b);
1508         return skb->len;
1509 }
1510
1511 struct tcf_action_net {
1512         struct rhashtable egdev_ht;
1513 };
1514
1515 static unsigned int tcf_action_net_id;
1516
1517 struct tcf_action_egdev_cb {
1518         struct list_head list;
1519         tc_setup_cb_t *cb;
1520         void *cb_priv;
1521 };
1522
1523 struct tcf_action_egdev {
1524         struct rhash_head ht_node;
1525         const struct net_device *dev;
1526         unsigned int refcnt;
1527         struct list_head cb_list;
1528 };
1529
1530 static const struct rhashtable_params tcf_action_egdev_ht_params = {
1531         .key_offset = offsetof(struct tcf_action_egdev, dev),
1532         .head_offset = offsetof(struct tcf_action_egdev, ht_node),
1533         .key_len = sizeof(const struct net_device *),
1534 };
1535
1536 static struct tcf_action_egdev *
1537 tcf_action_egdev_lookup(const struct net_device *dev)
1538 {
1539         struct net *net = dev_net(dev);
1540         struct tcf_action_net *tan = net_generic(net, tcf_action_net_id);
1541
1542         return rhashtable_lookup_fast(&tan->egdev_ht, &dev,
1543                                       tcf_action_egdev_ht_params);
1544 }
1545
1546 static struct tcf_action_egdev *
1547 tcf_action_egdev_get(const struct net_device *dev)
1548 {
1549         struct tcf_action_egdev *egdev;
1550         struct tcf_action_net *tan;
1551
1552         egdev = tcf_action_egdev_lookup(dev);
1553         if (egdev)
1554                 goto inc_ref;
1555
1556         egdev = kzalloc(sizeof(*egdev), GFP_KERNEL);
1557         if (!egdev)
1558                 return NULL;
1559         INIT_LIST_HEAD(&egdev->cb_list);
1560         egdev->dev = dev;
1561         tan = net_generic(dev_net(dev), tcf_action_net_id);
1562         rhashtable_insert_fast(&tan->egdev_ht, &egdev->ht_node,
1563                                tcf_action_egdev_ht_params);
1564
1565 inc_ref:
1566         egdev->refcnt++;
1567         return egdev;
1568 }
1569
1570 static void tcf_action_egdev_put(struct tcf_action_egdev *egdev)
1571 {
1572         struct tcf_action_net *tan;
1573
1574         if (--egdev->refcnt)
1575                 return;
1576         tan = net_generic(dev_net(egdev->dev), tcf_action_net_id);
1577         rhashtable_remove_fast(&tan->egdev_ht, &egdev->ht_node,
1578                                tcf_action_egdev_ht_params);
1579         kfree(egdev);
1580 }
1581
1582 static struct tcf_action_egdev_cb *
1583 tcf_action_egdev_cb_lookup(struct tcf_action_egdev *egdev,
1584                            tc_setup_cb_t *cb, void *cb_priv)
1585 {
1586         struct tcf_action_egdev_cb *egdev_cb;
1587
1588         list_for_each_entry(egdev_cb, &egdev->cb_list, list)
1589                 if (egdev_cb->cb == cb && egdev_cb->cb_priv == cb_priv)
1590                         return egdev_cb;
1591         return NULL;
1592 }
1593
1594 static int tcf_action_egdev_cb_call(struct tcf_action_egdev *egdev,
1595                                     enum tc_setup_type type,
1596                                     void *type_data, bool err_stop)
1597 {
1598         struct tcf_action_egdev_cb *egdev_cb;
1599         int ok_count = 0;
1600         int err;
1601
1602         list_for_each_entry(egdev_cb, &egdev->cb_list, list) {
1603                 err = egdev_cb->cb(type, type_data, egdev_cb->cb_priv);
1604                 if (err) {
1605                         if (err_stop)
1606                                 return err;
1607                 } else {
1608                         ok_count++;
1609                 }
1610         }
1611         return ok_count;
1612 }
1613
1614 static int tcf_action_egdev_cb_add(struct tcf_action_egdev *egdev,
1615                                    tc_setup_cb_t *cb, void *cb_priv)
1616 {
1617         struct tcf_action_egdev_cb *egdev_cb;
1618
1619         egdev_cb = tcf_action_egdev_cb_lookup(egdev, cb, cb_priv);
1620         if (WARN_ON(egdev_cb))
1621                 return -EEXIST;
1622         egdev_cb = kzalloc(sizeof(*egdev_cb), GFP_KERNEL);
1623         if (!egdev_cb)
1624                 return -ENOMEM;
1625         egdev_cb->cb = cb;
1626         egdev_cb->cb_priv = cb_priv;
1627         list_add(&egdev_cb->list, &egdev->cb_list);
1628         return 0;
1629 }
1630
1631 static void tcf_action_egdev_cb_del(struct tcf_action_egdev *egdev,
1632                                     tc_setup_cb_t *cb, void *cb_priv)
1633 {
1634         struct tcf_action_egdev_cb *egdev_cb;
1635
1636         egdev_cb = tcf_action_egdev_cb_lookup(egdev, cb, cb_priv);
1637         if (WARN_ON(!egdev_cb))
1638                 return;
1639         list_del(&egdev_cb->list);
1640         kfree(egdev_cb);
1641 }
1642
1643 static int __tc_setup_cb_egdev_register(const struct net_device *dev,
1644                                         tc_setup_cb_t *cb, void *cb_priv)
1645 {
1646         struct tcf_action_egdev *egdev = tcf_action_egdev_get(dev);
1647         int err;
1648
1649         if (!egdev)
1650                 return -ENOMEM;
1651         err = tcf_action_egdev_cb_add(egdev, cb, cb_priv);
1652         if (err)
1653                 goto err_cb_add;
1654         return 0;
1655
1656 err_cb_add:
1657         tcf_action_egdev_put(egdev);
1658         return err;
1659 }
1660 int tc_setup_cb_egdev_register(const struct net_device *dev,
1661                                tc_setup_cb_t *cb, void *cb_priv)
1662 {
1663         int err;
1664
1665         rtnl_lock();
1666         err = __tc_setup_cb_egdev_register(dev, cb, cb_priv);
1667         rtnl_unlock();
1668         return err;
1669 }
1670 EXPORT_SYMBOL_GPL(tc_setup_cb_egdev_register);
1671
1672 static void __tc_setup_cb_egdev_unregister(const struct net_device *dev,
1673                                            tc_setup_cb_t *cb, void *cb_priv)
1674 {
1675         struct tcf_action_egdev *egdev = tcf_action_egdev_lookup(dev);
1676
1677         if (WARN_ON(!egdev))
1678                 return;
1679         tcf_action_egdev_cb_del(egdev, cb, cb_priv);
1680         tcf_action_egdev_put(egdev);
1681 }
1682 void tc_setup_cb_egdev_unregister(const struct net_device *dev,
1683                                   tc_setup_cb_t *cb, void *cb_priv)
1684 {
1685         rtnl_lock();
1686         __tc_setup_cb_egdev_unregister(dev, cb, cb_priv);
1687         rtnl_unlock();
1688 }
1689 EXPORT_SYMBOL_GPL(tc_setup_cb_egdev_unregister);
1690
1691 int tc_setup_cb_egdev_call(const struct net_device *dev,
1692                            enum tc_setup_type type, void *type_data,
1693                            bool err_stop)
1694 {
1695         struct tcf_action_egdev *egdev = tcf_action_egdev_lookup(dev);
1696
1697         if (!egdev)
1698                 return 0;
1699         return tcf_action_egdev_cb_call(egdev, type, type_data, err_stop);
1700 }
1701 EXPORT_SYMBOL_GPL(tc_setup_cb_egdev_call);
1702
1703 static __net_init int tcf_action_net_init(struct net *net)
1704 {
1705         struct tcf_action_net *tan = net_generic(net, tcf_action_net_id);
1706
1707         return rhashtable_init(&tan->egdev_ht, &tcf_action_egdev_ht_params);
1708 }
1709
1710 static void __net_exit tcf_action_net_exit(struct net *net)
1711 {
1712         struct tcf_action_net *tan = net_generic(net, tcf_action_net_id);
1713
1714         rhashtable_destroy(&tan->egdev_ht);
1715 }
1716
1717 static struct pernet_operations tcf_action_net_ops = {
1718         .init = tcf_action_net_init,
1719         .exit = tcf_action_net_exit,
1720         .id = &tcf_action_net_id,
1721         .size = sizeof(struct tcf_action_net),
1722 };
1723
1724 static int __init tc_action_init(void)
1725 {
1726         int err;
1727
1728         err = register_pernet_subsys(&tcf_action_net_ops);
1729         if (err)
1730                 return err;
1731
1732         rtnl_register(PF_UNSPEC, RTM_NEWACTION, tc_ctl_action, NULL, 0);
1733         rtnl_register(PF_UNSPEC, RTM_DELACTION, tc_ctl_action, NULL, 0);
1734         rtnl_register(PF_UNSPEC, RTM_GETACTION, tc_ctl_action, tc_dump_action,
1735                       0);
1736
1737         return 0;
1738 }
1739
1740 subsys_initcall(tc_action_init);